From: Mike Linksvayer (ml@gondwanaland.com)
Date: Fri Apr 12 2002 - 05:00:53 MDT
On Thu, 2002-04-11 at 23:24, Eugen Leitl wrote:
> We have two bottlenecks: hardware, and what we can do with it. If there's
> no hardware, we clearly can't do anything. If there is hardware, but no
> methods, we don't look too hot, either. Only if there's both we are
> golden.
For any given hardware requirement it'll be here relatively/predictably
soon, which Moravec and Kurzweil make much hay of. The path to
removing/widening hardware bottleneck after bottleneck is exceedingly
clear. AFAICT the path to adequate software is nearly as murky as the
hardware path is clear.
> > than current hardware right now what could we do with it (along a
> > critical path to AI) that we _can't_ do with current hardware. NB: we
> > now have hardware 1000x faster than that available during the mid-80s AI
> > hypelet.
>
> Memory bandwidth went up by 10^3 since 1985? Wow. I must have missed more
> than just WW3.
I was referring to cpu speed. As measured by the STREAM benchmark
memory bandwidth has improved by 10^2, not shabby. See historical plots
at <http://www.cs.virginia.edu/stream/analyses.html>.
> As to hardware being irrelevant to real-world problem solving, you'll
> notice that the top500.org list growth shows no signs of slacking. If I
> indeed had a 1000-node cluster in a desktop volume I could go to places I
> couldn't with my current setup. If every Jane and Joe had such machines on
> the network, I would consider the SI scenario a lot more probable, as at
> least the hardware base would be closer to the requirements.
Of course hardware is relevant to real-world problem solving, that's the
economic engine that promotes investment in ever faster computers. How
much of that hardware is being used to tackle problems on a critical
path towards AI?
> The first use of a lot of big boxes is to find out how to use them
> efficiently. This is a very big, embarassingly parallel task. The more
> power you have to burn, the more stupid brute-force approaches we can
> afford, as smart approaches are limited by the human resource bottleneck,
> which is not getting any wider. Hardware helps a lot here.
Sure, though designing even stupid brute-force approaches isn't a walk
in the park if you're going to do anything other than generate excess
heat.
Mike Linksvayer
http://gondwanaland.com/ml/
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:13:26 MST